EGLNN:ENHANCED GRAPHLESS NEURAL NETWORK FOR IOT DATA STORAGE TRANSACTION SECURITY
Keywords: IoT, Ethereum, Graph Neural Network, Knowledge Distillation, Abnormal Detection
Abstract: With the rise of 5G and the IOT, the amount of data generated by IoT devices has exploded. Ethereum has become a secure tool for storing and trading IoT data due to its openness and tamper-proof nature. However, as Ethereum becomes more and more popular, the Ethereum platform has also become a hotbed for various types of cybercrimes, so ensuring the security of the Ethereum network is crucial. Recently, algorithms based on GNN have been seen as an effective way to detect abnormal nodes in the network. However, through analysis, this work finds that its original network structure is not optimal, directly applied to the existing GNN model with poor results. Meanwhile, it is understood that most of the current GNNs rely on the message-passing principle, which leads to slow model training and inference, and large model size. It is quite challenging to directly apply traditional GNN algorithms in industrial scenarios with limited space and high feedback time requirements. This study proposes a knowledge distillation-based algorithm called Enhanced Graph-Less Neural Network .EGLNN estimates more realistic graph structures through Bayesian graph structure estimator and solves the problem of large-scale GNN models being difficult to be widely applied in industry through the faculty-student distillation method.
Primary Area: unsupervised, self-supervised, semi-supervised, and supervised representation learning
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 10886
Loading